Peer Groups

Peer groups are clusters of organizations that share similar features. Understanding how your library peers perform is a powerful tool to support continuous improvement. 

There are countless ways libraries could be grouped together: budget, collection size, number of branches, number of staff, service area population, etc. However, up until now, there has not been a consistent or statistically verified way for libraries to make effective peer comparisons. Benchmark incorporates a set of 63 peer groups developed for PLA by the American Institutes for Research (AIR). Each public library (administrative entity) in the US has been assigned to a primary peer group. These peer groups were designed to be statistically valid and meaningful. The libraries within a peer group are not exactly the same, but they share similar characteristics. Those characteristics have been shown to impact the library’s outputs and outcomes.

This resource outlines the methods used to create the peer groups and characteristics of each group (see the Primary Peer Groups table below). Benchmark subscribers can see a list of the other libraries in their primary peer group in the Data Dashboards. 

Custom peer groups & other tools 

There may be instances when your library wants to select other variables as the basis for a peer group. Benchmark subscribers have the ability to create custom peer groups. View the tutorial on creating a custom peer group in the Resources. 

IMLS’s Library Search and Compare tool allows users to manually narrow all US libraries (administrative entities) down to a group of “peers” based on 10 variables. 

PLA’s Benchmark Briefings also rely on IMLS data and provide libraries with a snapshot of comparisons across key library characteristics. 

Establishing Primary Peer Groups 

In its ongoing investment to provide tools and resources to support libraries’ data analysis efforts, PLA identified the need to establish statistically valid and meaningful peer groups that libraries could use to inform their decision making. These peer groups were incorporated into Benchmark to help libraries easily compare their performance to their peers. 

To create the peer groups, AIR analyzed data from the IMLS Public Libraries Survey (PLS) and the U.S. Census Bureau to identify five characteristics that were most predictive of library outcomes (defined below): Region, Legal Basis, Locale, Percentage of population of the library’s census tract that is people of color (non-white), and the Percentage of the adult population of the library’s census tract that is college educated.

• The US is divided into nine regions: New England (CT ME MA NH RI VT); Mid East (DE DC MD NJ NY PA); Great Lakes (IL IN MI OH WI); Plains (IA KS MN MO NE ND SD); Southeast (AL AR FL GA KY LA MS NC SC TN VA WV); Southwest (AZ NM OK TX); Rocky Mountains (CO ID MT UT WY); Far West (AK CA HI NV OR WA); and Outlying Areas (AS GU MP PR VI). For the original peer group analysis, some categories were combined. Specifically, for region, New England was combined with the Mid East area, and the Southwest, Rocky Mountains, and Far West were combined. 
• Legal basis is the type of local government structure within which the entity functions. It reflects the state or local law, which authorizes the library. The IMLS Public Libraries Survey uses 9 legal basis categories. These were condensed down to 5 categories: county or city/county; municipal; library district; nonprofit; and other (includes multi-jurisdictional, Native American tribal government, school district, and other). 
• Locale is determined by the level of urbanization of the library’s service area. Locale types were collapsed into three groups: city, suburban, or town/rural. 
• Percentage people of color (non-white population) for the peer group creation was originally based on the population of the census tract in which the administrative entity is located. This data was drawn from the American Community Survey 5-year estimates using the Census Planning Database. 
• Percentage college educated for the peer group creation was based on the population of the census tract over age 25 that hold a bachelor’s degree or higher. This data was drawn from the American Community Survey 5-year estimates using the Census Planning Database. 

Based on this analysis, public libraries were grouped into 63 peer groups.

Libraries within a peer group do not necessarily share all the same characteristics. Rather, this means the libraries within a group are more similar to each other than to libraries in other groups. 

Methodology 

There are many ways to cluster libraries into groups with shared characteristics. In determining its methodology, AIR considered a number of factors: 

• The number of groups overall had to be large enough that there were recognizable differences between groups and similarities within groups. Libraries should recognize themselves as alike other libraries within their group. 
• The number of groups overall had to be small enough that there would be sufficient numbers of libraries within each group for effective benchmarking. 
• The peer groups’ definitions needed to be transparent and intuitive, so a library can identify why it falls in a particular group. 
• The peer groups had to be created based on data known for all public libraries. This meant that the variables used to create the peer groups had to either come from the PLS or from data sources that could be geographically linked to the PLS. 
• The peer groups had to be discrete and “re-creatable” using the PLS and other variables so that a new library not included in the original clustering process could later be assigned to a peer group. 

The original analysis was conducted using the FY2017 PLS data. The peer groups were updated when the FY2018 data was released. 

AIR considered several statistical clustering and classification methods to create the peer groups and settled on using recursive partitioning and regression trees. Broadly, recursive partitioning algorithms, often referred to as “tree-based” methods, are designed to automatically partition a dataset into clusters that differ with respect to some output or outcome measure. From a set of available variables, these algorithms identify the variable that best predicts the outcome, and then splits the dataset into “nodes'' defined by that variable. For example, if Locale were found to be the best available predictor for a given PLS outcome, one node might consist of libraries in cities and suburbs and the other node might consist of libraries in towns and rural areas. (The exact cutoff point is typically chosen to maximize differences between the nodes with respect to the outcome variable.) This process is then repeated within each node, leading to a successive set of splits that divide the data into a progressively smaller set of “child” nodes. The splitting continues until pre-defined stopping criteria (e.g., a minimum node size or measure of significant change) are met. 

Since PLA was interested in finding a classification system that would be robust across various measures (e.g., technology, salaries, or staffing measures), AIR used this method (regression trees) to develop classification models for multiple PLS measures, and then compared the variables that were most important across multiple measures. Because almost all PLS variables are at the administrative entity (AE) level, the analyses had to be conducted at that level, rather than the outlet, or branch, level. The measures (variables) that were chosen for the analysis were those that seemed likely to relate to technology variability in libraries (since peer group development was undertaken in conjunction with PLA’s the 2020 technology survey) or they were measures against which libraries may be interested in benchmarking themselves. 

Eleven dependent variables were selected from the PLS for use in the regression trees analysis (outlined in the table below).

Table 1: Dependent variables used in regression trees
Element Name	Element Description
TOTSTAFF	Total Paid FTE
BKVOL	Total Printed Material
TOTINCM	Total Operating Revenue
ELMATCIR	Circulation of electronic materials
VISITS	Total annual library visits
WIFISESS	Total annual wireless sessions provided by the library wireless service
TOTATTEN	Total audience at all library programs
TOTCIR	Total annual circulation transactions
TOTPRO	Total library programs
REFERENC	Total annual reference transactions
TOTOPEXP	Total operating expenditures

In addition, AIR selected a small subset of PLS and Census variables to use as predictor (independent) variables for the trees. These were the variables that would be used to define the peer group characteristics so they had to be both useful in identifying groups with similar values on the key measures and meaningful to libraries as measures by which they identify themselves. The Census variables came from the American Community Survey 5-year estimates in the Census Planning Database.

Table 2: Independent variables used in regression trees
Source	Element Name	Element Description
PLS	C_LEGBAS	Legal Basis Code
PLS	C_ADMIN	Administrative Structure Code
PLS	HRS_OPEN	Total Public Service Hours-All Outlets
PLS	TOTSTAFF	Total Paid FTE
PLS	BKVOL	Total Print Materials
PLS	LOCALE_ADD	Size of Community and Proximity to Urban/Metro Area based on Address of AE
PLS	OBEREG	Bureau of Economic Analysis Region Code
PLS	EBOOK	Electronic Books
PLS	NUM_OUTLET	Number of stationary outlets per AE
PLS	TOT_SQ_FEET	Total Square Feet per AE
PLS	TOTOPEXP	Total operating expenditures
Census	pct_Hispanic_ACS_13_17	The percentage of the population that identify as of Hispanic or Latino origin.
Census	pct_NH_White_alone_ACS_13_17	The percentage of the population that indicate no Hispanic origin and their only race as "White".
Census	pct_NH_Blk_alone_ACS_13_17	The percentage of the population that indicate no Hispanic origin and their only race as Black or African American.
Census	pct_Othr_Lang_ACS_13_17	The percentage of the population aged 5 years and over that speaks a language other than English at home.
Census	pct_Not_HS_Grad_ACS_13_17	The percentage of the population aged 25 years and over that are not high school graduates and have not received a diploma or the equivalent.
Census	pct_College_ACS_13_17	The percentage of the population aged 25 years and over that have a college degree or higher.
Census	pct_Prs_Blw_Pov_Lev_ACS_13_17	The percentage of the population that are classified as below the poverty level given their total family or household income within the last year, family size, and family composition.
Census	pct_Civ_unemp_16p_ACS_13_17	The percentage of civilians ages 16 years and over in the labor force that are unemployed.

Once all regression trees were run, the most important variables were selected to cross-classify cases into stratification and profile cells. Peer groups were then collapsed to minimize the number of groups and create groups with enough libraries to be meaningful and useful for sampling.

The results of the analysis showed that the region in which the AE is located (OBEREG) was the predictor that influenced the most output measures. After that, legal basis (C_LEGBAS), locale (LOCALE_ADD), the percent of individuals in the Census tract identified as non-Hispanic white (PCT_NH_WHITE), and the percent of individuals in the Census tract with a college degree (PCT_COLLEGE) were the most prevalent predictors. Therefore, these five variables were selected to create the peer groups. Some categories were combined and the Census variables were collapsed into categories in order to reduce the number of cells in the cross-classification (these changes are identified on page 2 above).

The cross-classification of these variables then formed 675 unique groups (5 legal bases, 5 regions, 3 locales, 3 percent white groups, and 3 percent college educated groups). Frequency distributions of libraries within each of these 675 groups were examined in order to construct an appropriate smaller set of strata. 63 unique peer groups were formed with the goal of having approximately 100 libraries in each group. Groups were combined using different variables depending on the distribution of libraries. The final 63 peer groups that resulted from this process are outlined in the table below. 

Among the 63 peer groups, the average size (number of libraries in the group) is 140 and median is 131. The smallest peer group has 33 libraries in it, and the largest has 531. 

Please note: These numbers apply to the original primary peer groups based on IMLS PLS FY 2018 data; PLA staff have added new libraries established since that time to primary peer groups that best match their characteristics.

Primary Peer Groups (Table) 

The following table shows all 63 peer groups, with the coding for each of the five classification variables as well as the count of libraries. Your library’s peer group number is identified in the dynamic text at the top of the Summary dashboard, as shown in the anonymized screenshot below.

Benchmark subscribers see a table that identifies the other libraries in their primary peer group at the bottom of the Summary tab in the Data Dashboards. For more information, please review the tutorial on the Data Dashboards.

Peer  Group	Legal  Basis*	Region	Locale	%  White	% of College- Educated Pop.  over 25	Number  of Libraries
1	CC or CO	Any	City	Any	Any	92
2	CC or CO	Southeast	Suburb	Any	Any	99
3	CC or CO	New England, Mid East, Great Lakes, Plains, Southwest, Rocky Mountains, Far West	Suburb	Any	Any	86
4	CC or CO	New England, Mid East, Great Lakes, Plains	Town or Rural	Any	Any	135
5	CC or CO	Southeast	Town or Rural	Any	<=14.9	161
6	CC or CO	Southeast	Town or Rural	Any	>15	151
7	CC or CO	Southwest, Rocky Mountains, Far West	Town or Rural	Any	<=14.9	83
8	CC or CO	Southwest, Rocky Mountains, Far West	Town or Rural	Any	>15	183
9	CI	Southwest, Rocky Mountains, Far West	City	Any	Any	122
10	CI	New England, Mid East, Great Lakes, Plains, Southeast	City	Any	Any	150
11	CI	Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, Far West	Suburb	75- 94.9	>=35	135
12	CI	New England, Mid East	Suburb	75- 94.9	>=35	267
13	CI	Any	Suburb	>=95	Any	92
14	CI	New England, Mid East	Suburb	75- 94.9	<=14.9 15-34.9	98
15	CI	Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, Far West	Suburb	75- 94.9	<=14.9 15-34.9	127
16	CI	Any	Suburb	<=74. 9	<=14.9	115
17	CI	New England, Mid East, Great Lakes	Suburb	<=74. 9	15-34.9	131
18	CI	Plain, Southeast, Southwest, Rocky Mountains, Far West	Suburb	<=74. 9	15-34.9	113
19	CI	Any	Suburb	<=74. 9	>=35	161
20	CI	New England, Mid East	Town or Rural	75- 94.9	<=14.9 15-34.9	170
21	CI	New England, Mid East	Town or Rural	>=95	15-34.9	192
22	CI	New England, Mid East	Town or Rural	75- 94.9	>=35	111
23	CI	Any	Town or Rural	>=95	>=35	141
24	CI	Great Lakes	Town or Rural	75- 94.9	15-34.9	194
25	CI	Great Lakes	Town or Rural	>=95	15-34.9	172
26	CI	Plains	Town or Rural	75- 94.9	15-34.9	531
27	CI	Plains	Town or Rural	>=95	15-34.9	452
28	CI	Southwest, Rocky Mountains, Far West	Town or Rural	<=74. 9	<=14.9	162
29	CI	Southwest, Rocky Mountains, Far West	Town or Rural	<=74. 9	5-34.9 >=35	156
30	CI	Southeast, Southwest, Rocky Mountains, Far West	Town or Rural	75- 94.9	15-34.9	241
31	CI	New England, Mid East, Great Lakes, Plains, Southeast	Town or Rural	<=74. 9	Any	156
32	CI	Great Lakes, Plains	Town or Rural	75- 94.9	<=14.9	165
33	CI	Southeast, Southwest, Rocky Mountains, Far West	Town or Rural	75- 94.9	<=14.9	125
34	CI	New England, Mid East, Great Lakes	Town or Rural	>=95	<=14.9	111
35	CI	Plains, Southeast, Southwest, Rocky Mountains, Far West	Town or Rural	>=95	<=14.9	105
36a	CI	Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, Far West	Town or Rural	75- 94.9	>=35	84
36b	CI	Southeast, Southwest, Rocky Mountains, Far West	Town or Rural	>=95	15-34.9	33
37	LD	Any	City	Any	Any	53
38	LD	New England, Mid East	Suburb	Any	Any	126
39	LD	Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, Far West	Suburb	Any	<=14.9 15-34.9	158
40	LD	Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, Far West	Suburb	Any	>=35	97
41	LD	Great Lakes	Town or Rural	<=74. 9 75- 94.9	<=14.9	127
42	LD	Great Lakes	Town or Rural	<=74. 9 75- 94.9	15-34.9	143
43	LD	Great Lakes	Town or Rural	>=95	15-34.9	123
44	LD	New England, Mid East	Town or Rural	Any	<=14.9 15-34.9	61
45	LD	Great Lakes	Town or Rural	>=95	<=14.9	80
46	LD	Any	Town or Rural	Any	>=35	85
47	LD	Plains	Town or Rural	<=74. 9 75- 94.9	<=14.9 15-34.9	115
48	LD	Southeast	Town or Rural	Any	<=14.9 15-34.9	94
49	LD	Southwest, Rocky Mountains, Far West	Town or Rural	Any	<=14.9 15-34.9	134
50	MJ, NL, SD, or, OT	Any	City or Suburb	Any	Any	174
51	MJ, NL, SD, or, OT	Great Lakes	Town or Rural	Any	Any	136
52	MJ, NL, SD, or, OT	Southeast	Town or Rural	Any	Any	134
53	MJ, NL, SD, or, OT	New England, Mid East, Plains, Southwest, Rocky Mountains, Far West	Town or Rural	Any	Any	164
54	NP	New England, Mid East	Suburb	<=74. 9	Any	74
55	NP	New England, Mid East	Suburb	75- 94.9	<=14.9 15-34.9	118
56	NP	New England, Mid East	Suburb	75- 94.9	>=35	133
57	NP	New England, Mid East	Suburb	>=95	Any	61
58	NP	New England, Mid East	City	Any	Any	34
59	NP	New England, Mid East	Town or Rural	Any	<=14.9	115
60	NP	New England, Mid East	Town or Rural	<=74. 9 75- 94.9	15-34.9	198
61	NP	New England, Mid East	Town or Rural	>=95	15-34.9	248
62	NP	New England, Mid East	Town or Rural	Any	>=35	165
63	NP	Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, Far West	City, Suburb, Town, or Rural	Any	Any	167

*Legal Basis Codes 

• CC–City/County 
• CI–Municipal Government (city, town, or village) 
• CO–County/Parish 
• LD–Library District 
• MJ–Multi-jurisdictional 
• NL–Native American Tribal Government 
• NP–Non-profit Association or Agency 
• SD–School District 
• OT–Other